Now that ESXi 5 has exposed the "multiple vCores per vSocket" settings in the New VM wizard, I'm wondering if there are any guidelines for the use of this feature. I know that in terms of the number of physical cores a VM can use, there's no difference between 4 vSockets, each with 1 vCore and 1 vSocket, with 4 vCores: In both cases, there are 4 vCores in the VM, and thus up to 4 of the host cores can be used to execute workloads. And I understand that condensing those vCores in to fewer vSockets (a greater than 1:1 vCore:vSocket ratio) can help with licensing of some software: for example, Server 2003 Standard which can only operate on two sockets at a time, but can take advantage of multiple cores per socket.
But what I can't find (and apologies if I'm just missing it when I google and search the forums) is whether the vCore/vSocket layout makes a difference for any other reasons. For example, if I create a VM with 1 vSocket, and configure that vSocket with 2 vCores, how does VMware schedule the underlying pCores? Does it have to schedule all the vCores of a single vSocket using only the pCores of a single pSocket? Or can VMware schedule those vCores on any host pCore, regardless of where that core is? Does it have a preference as far as keeping scheduling pCores on the same pSocket when servicing a mulit-vCore vSocket? Technically, cores on the same socket would be able to exchange information slightly faster in some situations, so there might be a slight advantage there? Maybe?
Additionally, what is the best practice in cases where I don't have a need for fewer vSockets, or where I'm not using software that has socket/core restrictions? I notice that converting a version 7 VM with n vCPUs results in a version 8 VM with n vSockets, each with a single vCore. So is the best practice to continue going with "single core" vSockets?
Thanks in advance,
Andrew
